#Trial API's response is 15 seconds. Billing Account API's response is 75 seconds.

15 messages · Page 1 of 1 (latest)

coral forge
#

I'm using a Trial API to develop a simple text generation tool. And to generate 500-700 tokens - it took 15 seconds.

I later switched to a Billing Account because I wanted to take the project live - and by just doing that, it is now taking 60-75 seconds per generation.

I haven't changed a single word in the codebase. And the project is still running locally. So by simply swapping the API keys, the response time is incredibly worse now.

What do I do? Who do I contact? I've tried to get in contact with OpenAI using their chatbot on help.openai.com - and it just ends the conversation with a "we'll get in contact soon". Is this a common issue?

coral forge
#

Any clue..?

hushed mirage
#

The API is much slower than the website. Odd that a free trial key is faster. I only have a billing account now to test

coral forge
hushed mirage
coral forge
#

Yeah.. In comparison of other models - text-davinci-002 and curie-001 are pretty fast. But again, I can't put out a tool in production with these models, they are just not.. as good.

grizzled pumice
#

VenomM it seemed this morning that when I used a different user property in the request that it worked normally again without the delay or ratelimiterror

#

So maybe they are rate limiting the API calls based on that field now

#

Would be good to know what that rate limit is

#

So I can warn the users

coral forge
rustic orchid
#

For past few days my app's latency has ranged from about 15 seconds to over 80 seconds per completion using gtp-4 /chat/complations endpoint. This with and without user parameter. Not rate limited that I'm aware of...still testing and only doing very low volume of usage. Prompt/completion size doesn't matter. Have written to support to see if there are any account related issues.

grizzled pumice
#

Okay

#

It's a parameter like messages and temperature

#

It's just a theory